SBIR Phase I: Computer Vision for Merchandizing Forest Products

这个小型企业创新研究(SBIR)第一阶段项目的更广泛影响将来自于将尖端人工智能技术应用于伐木行业。伐木(木材采伐)是生产纸包装、保健和卫生产品、服装纤维，甚至是我们技术设备中使用的树脂的第一步。林产工业雇佣了近百万人，每年为美国经济贡献数千亿美元。然而，近几十年来，伐木领域几乎没有什么技术创新。该项目旨在研究和开发计算机视觉技术，以增强人们对收获的树木进行准确分级的能力。由此产生的技术可能会增加农村地区非熟练工人的就业机会，确保有效和高效地利用收获的树木，并增加木材供应链中数千家小企业的收入。这个小型企业创新研究(SBIR)第一阶段项目旨在证明计算机视觉系统的可行性，以增强人机操作员在分级和分类原木任务中的技能。在该项目期间，将建立一个适用于特定领域的数据集，将创建和培训一系列新的计算机视觉模型，并将在一个偏远环境中部署一个完全集成的原型。为了实现这些目标，该公司将研究使用自我监督学习来加快特定领域数据集的创建，以及可适应的模型链和模型压缩，以在日志记录站点实现有效的推理(即，不需要云计算资源)。该公司还将创建新的方法来确定特定的感兴趣对象(如缺陷)，并评估每个原木的等级。如果成功，该项目将展示一种计算机视觉系统，该系统能够识别和定位感兴趣的特定日志，跟踪日志，评估每个日志的尺寸，定位日志上的缺陷，准确确定每个日志的等级，并向机器操作员提供视觉反馈-所有这些都在操作员的简短决定时间框架内。该奖项反映了NSF的法定使命，并通过使用基金会的智力优势和更广泛的影响审查标准进行评估，被认为值得支持。

The broader impact of this Small Business Innovation Research (SBIR) Phase I project will result from applying leading-edge artificial intelligence technology to the logging industry. Logging (wood harvesting) is the first step in the production of paper packaging, health and hygiene products, clothing fibers, and even resins that are used in our technology devices. The forest products industry employs nearly one million people and contributes hundreds of billions of dollars to the US economy each year. Yet, there has been very little technological innovation in logging in recent decades. This project aims to research and develop computer vision technology that will augment a person’s ability to grade harvested trees accurately. The resulting technology may increase the number of jobs available to unskilled workers in rural areas, ensure effective and efficient utilization of harvested trees, and increase revenues for thousands of small businesses in the wood supply chain.This Small Business Innovation Research (SBIR) Phase I project aims to demonstrate the feasibility of a computer vision system to augment the skills of human-machine operators in the tasks of grading and sorting logs. During the project, a suitable domain-specific dataset will be established, a new chain of computer vision models will be created and trained, and a fully integrated prototype will be deployed in a remote environment. To achieve these goals, the company will research the use of self-supervised learning to expedite the creation of a domain-specific dataset, along with adaptable chains of models and model compression to enable efficient inference at the logging site (i.e., without the need for cloud computing resources). The company will also create new methods for determining specific objects of interest (such as defects) and assessing the grade of each log. If successful, the project will demonstrate a computer vision system that is able to identify and locate specific logs of interest, track the logs, assess each log’s dimensions, locate defects on the logs, accurately determine a grade for each log, and give visual feedback to a machine operator – all within the operator’s brief decision timeframe.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.